Slashdot Mirror
AMD Trinity A10-4600M Processor Launched, Tested
MojoKid writes "AMD lifted the veil on their new Trinity A-Series mobile processor architecture today. Trinity has been reported as offering much-needed CPU performance enhancements in IPC (Instructions Per Cycle) but also more of AMD's strength in gaming and multimedia horsepower, with an enhanced second generation integrated Radeon HD graphics engine. AMD's A10-4600M quad-core chip is comprised of 1.3B transistors with a CPU base core clock of 2.3GHz and Turbo Core speeds of up to 3.2GHz. The on-board Radeon HD 7660G graphics core is comprised of 384 Radeon Stream Processor cores clocked at 497MHz base and 686Mhz Turbo. In the benchmarks, AMD's new Trinity A10 chip outpaces Intel's Ivy Bridge for gaming but can't hold a candle to it for standard compute workloads or video transcoding."
They should do that the other way round.
I've seen a lot of reviews of various laptops that have missed the most important metric in this competition - Price!
What's been common in all reviews is that the only the very top end Intel "integrated" (No separate, discreet GPU) solutions have been competitive to the new fusion products. We're talking mobile i7s. I don't know if you've priced laptops lately, but the i7's are only found in expensive, high end systems.
The fusion APUs are nowhere near that expensive. Price wise, they should be compared to i3s or "pentium" mobile cpus.. Where they will win quite handily!
It turns out that AMD's 'APU' solutions have been very popular with low end device makers and AMD sells them by the boat load. What's impressed me, however, is how much intel has improved their GPU in ivy bridge. It's always been garbage before, but now it's starting to be something you could call 'low end'.
And they are, as long as you understand that they are not trying to compete at the level of a core i7. If you need that kind of x86 performance you have one choice, Intel, and you will pay their premium tier pricing to get it... AMD stumbled with the release of the FX series, hopefully as they move forward they will remain competitive.
That's really all that matters. I've always been and AMD fan but If they can't pull out the same performance for less or equal price, they're done.
IMO, the Trinity is a truly compelling offering from AMD, after a long long time. Yes, it trades lower CPU int/float performance for higher GPU performance when compared to Ivy Bridge, but this tradeoff makes it a very attractive choice for someone who wants a cheap to mid-priced laptop that gives you decent performance and decent battery life while still letting you play the latest bunch of games in low-def setting. Its hitting the sweet spot for laptops as far as I am concerned. I'm also fairly sure it will be priced about a hundred bucks cheaper than a comparable Ivy Bridge - that's how AMD has traditionally competed. Hats off to AMD fror getting their CPU performance to somewhat competitive levels while still maintaining the lead against the massively improved GPU of the Ivy Bridge. All this while they're still at 32nm while Ivy Bridge is at 22nm.
Having said that, what I am equally excited about is the hope that Intel will come up with Bay Trail, their 22nm Atom that I strongly suspect will feature a similar graphics core that is there in Ivy Bridge. Intel has always led with performance and stability, not with power efficiency and price, so they need to create something that genuinely beats the ARM design, at least in the tablet space if not in the cellphone space.
They used to be able to beat Intel in the Athlon days. Now they are hopelessly far behind and dumping huge hot graphics cores into their chips putting them further and further behind. Focus on cheap compute with unlocking cores AMD. Not stupid graphics cores which do nothing for the CPU. A 16 core phenom ii at $100 will sell much better than this insane graphics + cpu crap.
That is pretty much the exact opposite of a good plan for AMD(as much as I would like cheap compute...) Since Intel has a process advantage, and presently has a superior x86 compute core architecture, they can almost certainly beat AMD on production cost for chips of a given level of punch. Trying to compete on price with somebody kicking out chips a process node ahead of you just isn't a good plan. Unless they really fuck it up, or their yields tank horribly or similar, they'll be able to beat you on production cost every time. Intel has little to gain by cutting its own margins in order to chase AMD down a hole(since lower margins are bad, and killing AMD would mean becoming antitrust scrutiny case #1 for the indefinite future...); but there isn't any architectural barrier to their doing so.
Since Intel has comparatively worthless GPU designs, tacking GPUs onto CPU dice is a way for AMD to offer something that Intel cannot(and at a price lower than a discrete CPU + discrete GPU without totally cutting their own throat), and also happens to go well with today's enthusiasm for laptops and all-in-ones. They have a second niche, much more directly focused on price, in compute-light, memory-heavy server applications(since you can populate your sockets with AMD CPUs for less and the number of DIMMs you get is roughly proportional to the number of sockets you have active); but competing on price isn't good for your margins.
With an inferior process and a weaker x86 design, gunning directly for the compute performance crown would just be asking for a whupping from Intel.
> Ivy Bridge and Llano actually ended up 'tied
Yes, but Llano is the *old* AMD processor ;-) Check the reviews for performance of a HD 4000 vs a Trinity.
Speaking of all-in-ones, an all-in-one AMD chip would be a dandy basis for a games console. If not one from Microsoft (who has no particular need for x86) then it would perhaps be a good match for Valve. Public distaste for Sony is at an all-time high, but is it enough to unseat them? etc etc.
if I could have a 16 core phenom ii, though, that would be pretty awesome. I could drop it right into my current machine. I'd pay $100 for even eight cores, though, let alone sixteen.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Appreciating competition is not mutually exclusive with being critical of the competitions quality.
Log the fuck in so I can be sure I'm talking to the same moron who posted "Since ivy bridge intels GPU now equals anything from AMD."
Don't know something? Look it up. Still don't know? Then ask.
All of AMD's A-series processors make a great HTPC platform. Its been over a year now with Intel not offering any real competition at all in this segment once price is factored in. You can trivially get a full 65W A-series HTPC box up and running for under $150 with lots of headroom (thats the price I would quote to friends/coworkers and pocket the difference as labor costs.) The higher end A-series (100W) are only necessary if you are gaming.
' Some might say that Intel Atom solutions are price competitive with the A-series but the Atom solutions, just like AMD's low powered E-series lineup, really only works well for HTPC as long as 100% of your needed video codecs use GPU acceleration. If the Atom is good enough, then an E-series of the same price will be a bit better as well. Its hard to guarantee that all the codecs that you will be using will be GPU accelerated, especially so if you are stacked up on a Linux distro, so the E-series and Atoms are not really a solution that I recommend.
"His name was James Damore."
So the "any" is clearly restricted to the GPUs embedded in the CPUs not discrete GPUs in huge cards.
Okay - So talking about AMD having 384 stream processors per die in the 7660, vs... 16 for Intel in the HD4000. Not even the same game, never mind the same ballpark.
Sorry, but AMD wins this round. And although the average Joe hasn't yet realized it, the "number of cores" war has turned a corner, in that the CPU has already started serving merely as an "overseer" of massive numbers of GPU SPs/CUs. If you do, specifically and exclusively, transaction processing - The CPU still wins. In scientific computing, cryptography, signal analysis, physical simulations, CAD, and yes, even gaming - No one cares if you have a 12-way Xeon or an AMD Geode, it matters that you have an AMD 59/69/79xx (and yes, I do mean AMD - despite their overall gaming performance, for GPU computing, even NVidia doesn't even come close, though the uber-expensive Tesla does at least get to share the playing field).
/ Note that the recent Slashdot article on media transcoding dealt specifically with mass-market solutions using hacked-up shader routines, not optimized OpenCL kernels.
They have a second niche, much more directly focused on price, in compute-light, memory-heavy server applications(since you can populate your sockets with AMD CPUs for less and the number of DIMMs you get is roughly proportional to the number of sockets you have active)
I haven't tried AMD's latest server machines, but if they are even 1/2 as good as the old, ones they are a _MUCH_ better deal. My 6 !! year old DL585G2 is actually faster on every single thing it gets used for than the much newer westmere machines we have been buying. The problem is that intel is charging an absolute fortune for chips clocked fast, so we end up with 1.8 or 2.2Ghz westmere machines, and their single thread performance is abysmal compared to the much older 3.2Ghz AMD machine. Our application scales nicely, but quickly becomes IO bound, so both machines basically get the same throughput, but the AMD machine has much lower overall latency. This results in it actually getting much better benchmarks in our tests.
So, in theory we could get an intel that kicks the crap out of the AMD machine, but its going to cost us 5x as much (from ~$5k to ~$25k). So we buy the cheap ones, and they get their ass handed to them by a 6 year old machine that cost $5k when it was new.